Universal Potential: M3GNet-UP-2022

Note: This example requires AMS2023 or later.

This example illustrates how set up an AMS calculation using the M3GNet 2022 Universal Potential from the MLPotential Engine.

To follow along, either

Worked Example

Requirements

The package m3net can be installed with amspackages

Purpose

Use the M3GNet ML potential with AMS.

Initialization

from scm.plams import *

# this line is not required in AMS2025+
init()

Setup and run job

mol = from_smiles("O")
mol.lattice = [
    [
        3.0,
        0.0,
        0.0,
    ],
    [0.0, 3.0, 0.0],
    [0.0, 0.0, 3.0],
]

s = Settings()
s.runscript.nproc = 1
s.input.ams.task = "GeometryOptimization"
s.input.ams.GeometryOptimization.Convergence.Gradients = 0.01  # hartree/ang

s.input.MLPotential.Model = "M3GNet-UP-2022"
# If you have trained a custom M3GNet model yourself, you can use:
# s.input.MLPotential.Model = 'Custom'
# s.input.MLPotential.Backend = 'm3gnet'
# s.input.MLPotential.ParameterDir = '<path to directory containing the M3GNet model>'

job = AMSJob(settings=s, molecule=mol, name="ams_with_m3gnet")
job.run()

energy = job.results.get_energy(unit="eV")
print(f"M3GNet: final energy {energy:.3f} eV")

[04.03|15:36:21] JOB ams_with_m3gnet STARTED
[04.03|15:36:21] Renaming job ams_with_m3gnet to ams_with_m3gnet.002
[04.03|15:36:21] Job ams_with_m3gnet.002 previously run as ams_with_m3gnet, using old results
[04.03|15:36:21] JOB ams_with_m3gnet.002 COPIED
M3GNet: final energy -14.077 eV

Complete Python code

#!/usr/bin/env amspython
# coding: utf-8

# ## Requirements
#
# The package m3net can be installed with `amspackages`

# ## Purpose
#
# Use the M3GNet ML potential with AMS.

# ## Initialization

from scm.plams import *

# this line is not required in AMS2025+
init()


# ## Setup and run job

mol = from_smiles("O")
mol.lattice = [
    [
        3.0,
        0.0,
        0.0,
    ],
    [0.0, 3.0, 0.0],
    [0.0, 0.0, 3.0],
]

s = Settings()
s.runscript.nproc = 1
s.input.ams.task = "GeometryOptimization"
s.input.ams.GeometryOptimization.Convergence.Gradients = 0.01  # hartree/ang

s.input.MLPotential.Model = "M3GNet-UP-2022"
# If you have trained a custom M3GNet model yourself, you can use:
# s.input.MLPotential.Model = 'Custom'
# s.input.MLPotential.Backend = 'm3gnet'
# s.input.MLPotential.ParameterDir = '<path to directory containing the M3GNet model>'

job = AMSJob(settings=s, molecule=mol, name="ams_with_m3gnet")
job.run()

energy = job.results.get_energy(unit="eV")
print(f"M3GNet: final energy {energy:.3f} eV")